cross-lagged networks
Definition
Cross-lagged networks refers to a network modeling approach for longitudinal panel data in which individual observed items, such as symptoms, attitudes, or behaviors, are modeled as directly predicting one another across discrete measurement occasions rather than being subsumed under latent constructs. The model estimates auto-regressive and cross-lagged pathways among items using a combination of regularized regression and structural equation modeling, producing a directed network that characterizes how components of psychological constructs influence each other over time. Applied to students' commitment to school and self-esteem, the approach can identify specific predictive pathways, such as a link between perceived school performance and feelings of self-worth, that latent variable models would obscure by aggregating items into scale scores. Estimation proceeds in two steps using l1-regularized regression followed by structural equation modeling, a procedure shown through simulation to recover the true network structure effectively, particularly when relationships change across measurement occasions.
Sources: Wysocki et al. (2025)
Related Terms
- cross-lagged panel design (1 shared article)
- network models (1 shared article)
Applications
Cross-lagged Networks and Latent Variable Models
Cross-lagged networks are explicitly positioned as an alternative to latent variable cross-lagged panel models, which represent constructs as unobserved common causes of observed items and estimate predictive effects at the construct level. Where latent variable models treat item covariance as a reflection of a single underlying attribute, cross-lagged networks hypothesize direct relations among items, a distinction with consequences for the specificity of hypotheses that can be generated and tested.
Sources: Wysocki et al. (2025)
Cross-lagged Networks and Experience Sampling Method Data
Cross-lagged networks address a gap in the network modeling toolkit by extending item-level longitudinal modeling to panel designs, which collect data at a small number of widely spaced occasions, in contrast to vector autoregressive approaches suited to intensive longitudinal data gathered via the Experience Sampling Method. The panel network model therefore serves researchers who have multi-wave developmental data but lack the dense time series required by graphical VAR estimation.
Sources: Wysocki et al. (2025)
